cannam@86:
cannam@86: Terminology discussed in this document is based on common terminology cannam@86: associated with contemporary codecs such as MPEG I audio layer 3 cannam@86: (mp3). However, some differences in terminology are useful in the cannam@86: context of Vorbis as Vorbis functions somewhat differently than most cannam@86: current formats. For clarity, then, we describe a common terminology cannam@86: for discussion of Vorbis's and other formats' audio quality.
cannam@86: cannam@86:Objective fidelity is a measure, based on a computable, cannam@86: mechanical metric, of how carefully an output matches an input. For cannam@86: example, a stereo amplifier may claim to introduce less that .01% cannam@86: total harmonic distortion when amplifying an input signal; this claim cannam@86: is easy to verify given proper equipment, and any number of testers are cannam@86: likely to arrive at the same, exact results. One need not listen to cannam@86: the equipment to make this measurement.
cannam@86: cannam@86:However, given two amplifiers with identical, verifiable objective cannam@86: specifications, listeners may strongly prefer the sound quality of one cannam@86: over the other. This is actually the case in the decades old debate cannam@86: [some would say jihad] among audiophiles involving vacuum tube versus cannam@86: solid state amplifiers. There are people who can tell the difference, cannam@86: and strongly prefer one over the other despite seemingly identical, cannam@86: measurable quality. This preference is subjective and cannam@86: difficult to measure but nonetheless real.
cannam@86: cannam@86:Individual elements of subjective differences often can be qualified, cannam@86: but overall subjective quality generally is not measurable. Different cannam@86: observers are likely to disagree on the exact results of a subjective cannam@86: test as each observer's perspective differs. When measuring cannam@86: subjective qualities, the best one can hope for is average, empirical cannam@86: results that show statistical significance across a group.
cannam@86: cannam@86:Perceptual codecs are most concerned with subjective, not objective, cannam@86: quality. This is why evaluating a perceptual codec via distortion cannam@86: measures and sonograms alone is useless; these objective measures may cannam@86: provide insight into the quality or functioning of a codec, but cannot cannam@86: answer the much squishier subjective question, "Does it sound cannam@86: good?". The tube amplifier example is perhaps not the best as very few cannam@86: people can hear, or care to hear, the minute differences between tubes cannam@86: and transistors, whereas the subjective differences in perceptual cannam@86: codecs tend to be quite large even when objective differences are cannam@86: not.
cannam@86: cannam@86:Audio artifacts and loss of fidelity or more simply cannam@86: put, audio differences are not the same thing.
cannam@86: cannam@86:A loss of fidelity implies differences between the perceived input and cannam@86: output signal; it does not necessarily imply that the differences in cannam@86: output are displeasing or that the output sounds poor (although this cannam@86: is often the case). Tube amplifiers are not higher fidelity cannam@86: than modern solid state and digital systems. They simply produce a cannam@86: form of distortion and coloring that is either unnoticeable or actually cannam@86: pleasing to many ears.
cannam@86: cannam@86:As compared to an original signal using hard metrics, all perceptual cannam@86: codecs [ASPEC, ATRAC, MP3, WMA, AAC, TwinVQ, AC3 and Vorbis included] cannam@86: lose objective fidelity in order to reduce bitrate. This is fact. The cannam@86: idea is to lose fidelity in ways that cannot be perceived. However, cannam@86: most current streaming applications demand bitrates lower than what cannam@86: can be achieved by sacrificing only objective fidelity; this is also cannam@86: fact, despite whatever various company press releases might claim. cannam@86: Subjective fidelity eventually must suffer in one way or another.
cannam@86: cannam@86:The goal is to choose the best possible tradeoff such that the cannam@86: fidelity loss is graceful and not obviously noticeable. Most listeners cannam@86: of FM radio do not realize how much lower fidelity that medium is as cannam@86: compared to compact discs or DAT. However, when compared directly to cannam@86: source material, the difference is obvious. A cassette tape is lower cannam@86: fidelity still, and yet the degradation, relatively speaking, is cannam@86: graceful and generally easy not to notice. Compare this graceful loss cannam@86: of quality to an average 44.1kHz stereo mp3 encoded at 80 or 96kbps. cannam@86: The mp3 might actually be higher objective fidelity but subjectively cannam@86: sounds much worse.
cannam@86: cannam@86:Thus, when a CODEC must sacrifice subjective quality in order cannam@86: to satisfy a user's requirements, the result should be a cannam@86: difference that is generally either difficult to notice cannam@86: without comparison, or easy to ignore. An artifact, on the cannam@86: other hand, is an element introduced into the output that is cannam@86: immediately noticeable, obviously foreign, and undesired. The famous cannam@86: 'underwater' or 'twinkling' effect synonymous with low bitrate (or cannam@86: poorly encoded) mp3 is an example of an artifact. This cannam@86: working definition differs slightly from common usage, but the coined cannam@86: distinction between differences and artifacts is useful for our cannam@86: discussion.
cannam@86: cannam@86:The goal, when it is absolutely necessary to sacrifice subjective cannam@86: fidelity, is obviously to strive for differences and not artifacts. cannam@86: The vast majority of codecs today fail at this task miserably, cannam@86: predictably, and regularly in one way or another. Avoiding such cannam@86: failures when it is necessary to sacrifice subjective quality is a cannam@86: fundamental design objective of Vorbis and that objective is reflected cannam@86: in Vorbis's design and tuning.
cannam@86: cannam@86: